Follow
Subscribe

NVIDIA Takes Drive PX2 Driverless Platform Testing to the Next Phase

Home > News > Content

【Summary】Its Drive PX2 autonomous driving platform, which is being used by cars participating in the highly anticipated Roborace circuit, utilizes a special form of deep learning, called convolutional neural network (CNN).

Original Michael Cheng    Dec 16, 2016 6:32 AM PT

Almost all driverless systems being tested today requires a set of clearly marked lanes and robust sensors to ensure the vehicle stays on its path. The problem with this framework is its strong reliance on well-developed infrastructure. When encountering broken roads without bold markings that are filled with small patches of grass (like some roadways in rural locations), autonomous vessels would have a very difficult time navigating the area.

NVIDIA's solution to such issues is very unique. Its Drive PX2 autonomous driving platform, which is being used by cars participating in the highly anticipated Roborace circuit, utilizes a special form of deep learning, called convolutional neural network (CNN). For developing self-driving capabilities, company engineers fed the CNN time-stamped videos to "teach" the system how to operate the vessel like a human. This technique has allowed cars using NVIDIA's self-driving platform to steer around locations without lanes.

The chipmaker's fixation is currently "on bringing high-level artificial intelligence into the autonomous driving game, especially supporting SAE Level 4 and 5," said Praveen Chandrasekar, mobility research manager at Frost & Sullivan.

 DRIVE+PX_illustration.jpg

Testing on California Roads

Earlier this month, the California Department of Motor Vehicles awarded NVIDIA a permit for the testing of driverless platforms in the state. It now joins a long list of car manufacturers that are rigorously developing their own driverless programs, including Tesla, Ford, GM, Faraday Future and Mercedes-Benz. The business has already been seen piloting its popular BB8 AI test vehicle around the Bay Area in San Francisco.

NVIDIA's move from a reputable graphics hardware manufacturer to a platform developer for self-driving cars could not have come at a better time. The company's deep expertise in parallel computing has allowed it to build an incredibly smart driverless system with robust problem-solving capabilities. Additionally, the AI-powered technology can also be applied to other platforms outside of autonomous cars, such as deep learning software toolkits for tech-savvy businesses with complex automation requirements.   

 BaiduWorld_JHH_logos.0.0.jpg

NVIDIA and Baidu

The chipmaker is not developing its self-diving products alone. In September, the brand announced a partnership with Baidu to build a platform for semi-autonomous cars. It is important to consider that both companies now have permits to test driverless vehicles in the state of California. According to a report from TechCrunch, the Chinese internet giant plans to use the experimental platform to launch its own commercial fleet.

In the collaboration, NVIDIA offers its cutting edge experience with AI systems, while Baidu provides expertise surrounding digital mapping. So far, the duo is aiming for SAE L3 autonomy, also known as "Conditional Automation" – which relies on a human driver for fallback performance. Together, the newly forged team will target OEMs interested in streamlining the development of autonomous vehicular components.

"It's likely the company will be developing the system to offer to other self-driving car manufacturers, as we can't see the graphics card maker developing its own fleet of autonomous vehicles," explained Joe Roberts from Trusted Reviews.

The biggest threat to the duo's self-driving project is Intel. The establishment is quickly scaling its efforts in the driverless sector, starting with a massive $250 million investment in self-driving technology and the formation of the Automated Driving Group (ADG). 

Prev                  Next
Writer's other posts
Comments:
    Related Content